The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill handles untrusted data from Salesforce record fields which could potentially contain malicious instructions.
Ingestion points: Data is retrieved via soql_query and rest_call as described in Phase 2, 3, and 4 of SKILL.md.
Boundary markers: Absent; the skill does not instruct the agent to wrap record data in delimiters or provide "ignore instructions" warnings.
Capability inventory: The skill has access to DML operations (sobject_dml), Tooling API operations, and Apex code execution (apex_execute) as specified in the tool mappings in references/execution-modes.md.
Sanitization: Basic normalization (trimming, casing, and stripping punctuation) is performed as part of the deduplication strategy, but no specific prompt injection sanitization is present.
[DYNAMIC_EXECUTION]: Instructions in references/execution-modes.md direct the agent to generate and execute local scripts (Python, shell, jq) for data processing and report generation based on metadata and data artifacts retrieved from the org.
[COMMAND_EXECUTION]: The skill utilizes shell commands and the Salesforce CLI tool (sf) to detect environment configuration, verify project metadata, and perform data operations.

sf-records